In this paper, we propose a new scoring method for local feature-based image retrieval. The proposed score is based on the ratio of the probability density function of an object model to that of background model, which is efficiently calculated via nearest neighbor density estimation. The effectiveness of the proposed method is confirmed in the experiments.

抄録全体を表示

This paper studies a video-to-panorama registration method for video sequences with zoom change. Conventional stitching methods have a problem of image combination including deep zoom scenes. Then, we propose a detection method of zoom scenes using optical flows, and an image selection method especially for zoom scenes for panoramic images. Experimental results show that our method can precisely create the whole panorama images.

抄録全体を表示

Expression intensity of each face image is obtained as value which humans sense. A method is proposed for estimation of a facial expression intensity close to a human sense thorough regression analyses between facial expression intensities and facial image vectors. Moreover, we also propose a method for facial expression recognition.

抄録全体を表示

We have developed a prototype system which enables SNS users to create CG based video content easily and provides communication that starts with the video content creation.

抄録全体を表示

Our goal is to let people experience and learn the nature of the ocean. We propose an algorithm for behavior generalization of fish school by the potential field method. The proposed incorporation of this method has potential to keep computational cost down, and generate various and adaptive behaviors in 3D-spaces.

抄録全体を表示

We propose a messaging system which achieves a function for easy way of replying to messages. Receivers of messages are able to reply to the sender of messages at one touch of a button on the touch screen which displays the application screen with the body of the received message and a few buttons.

抄録全体を表示

We have been studying the interactive panoramic video delivery method to aim at providing the highly realistic feeling for the user who is in the home as if he/she enjoyed watching in the event site, in which the whole event scene is distributed by a high resolution panorama video and the home user selects the viewing point and watches. We report the subjective evaluation results on realistic feeling for users by touch interface of Multi-Touch-Display linked to the TV set.

抄録全体を表示

It's important to continue moderate physical activity or exercise for prevention of lifestyle related disease. So, we propose "VR Trampoline" consisted of VR and trampoline exercise. This system shows the user image linked exercise. We made two applications. From this research, we confirmed more improvement of enjoyment and exercise motivation.

抄録全体を表示

This study compares impressions of images displayed with recognizable luminance-differences and with unrecognizable luminance-differences, in order to examine luminance-differences that are suitable for high-gradation display.

抄録全体を表示

This paper presents features for measuring congestion area in endoscopic images of internal hemorrhoids. The congestion is a symptom in the hemorrhoids and the features could be useful as variables in discrimination of the hemorrhoid cases. Experimental results using 80 cases showed ratio of the discrimination are more than 85 %.

抄録全体を表示

The SVM, which is used as a powerful tool for cancer/noncancer discrimination for biopsy images, has difficulty in the parameter tuning. To avoid this problem, we propose a discrimination method using a parameter-free sparse representation classifier. Experimental results show that the performance of the proposed method is comparable to that of SVM.

抄録全体を表示

A set of novel "person-independent deformable templates" (PIDTs) was developed for face tracking applications. After briefly describing the construction of the PIDTs, we describe their application to face detection at multiple poses, and present experimental results confirming the operation of the system.

抄録全体を表示

In this research, we propose a method for age-group classification from face images using Gabor features. We first made a comparison of each element of the Gabor features obtained on face images in terms of sensitivity to aging effect found on faces among different subjects, and conducted an age estimation experiment based on the selected features with higher sensitivity to aging.

抄録全体を表示

We defined feature vectors that represent the 3D coordinates of marker points attached on human face, which were measured by a motion capture system, and investigated discrimination capability of the features among several categories of facial expressions by classification experiments.

抄録全体を表示

We built a visual interface system in which automatic eye-direction classification robust against head pose variation was achieved by acquiring the 3D posture information of the face by means of a state-of-the-art head tracking software, faceAPI. We made a comparison between several feature extraction schemes for representing the local image pattern around the eyes in terms of validity in discriminating eye directions.

抄録全体を表示

In gaze tracking system, we propose the interactive technique of correcting measured value in real time. As a result of mounting a calibration interface, a possibility that very high-precision gaze measurement would be realizable was shown.

抄録全体を表示

We propose a generation method of music melodies for image slide-show based on visual features in the image. Experimental results show that consistency of generated melodies and transition effects in slide show is important for human impression.

抄録全体を表示

This paper presents the results of subjective evaluation of Music Slide Show System. There were some problems that the atmosphere of pictures used for the slideshow was different to that of lyrics of the music. We analyzed the results and improved the quality of system.

抄録全体を表示

We propose a movie search system allowing abstract queries caused by users' vague memories. The prototype system shows correct results matching the user's intention even if the user does not have a concrete query. We developed two sub-systems: a query expansion system based on semantic relations, and a search system using textual alignment recognition. We compared them with a TF-IDF based search algorithm, and showed that the rank of the correct answer is higher in our system than in the compared algorithm.

抄録全体を表示

We developed virtual reality content of sea fish for science education in primary school. The content was displayed on a cave-like multi-projection system to provide real scale images of fishes with high realistic sensation. We outline the content and describe the experimental results of subject's size and depth perception to the displayed fishes.

抄録全体を表示

One of the factors of simulator sickness is the absence of centrifugal force when taking turns. We proposed a simulator whose vision inclines in accordance with the centrifugal acceleration so as to induce inclination of driver's head. Outward inclination of outside view relieved simulator sickness, and improved the reality and the easiness of driving at curves in young drivers.

抄録全体を表示

The bright and wide field-of-view (FOV) displays of future TV systems can strongly stimulate human peripheral vision, which is sensitive to flicker. For temporal specification of future TV systems, we measured the critical fusion frequency (CFF) in 26 participants, using varying luminance, duty ratios, and FOVs.

抄録全体を表示

We estimated the volume of the color gamut of the Munsell Book of Color under daylight illuminants and blackbody radiators with different (correlated) color temperatures. We point out some limitations of the color appearance models used for the estimation.

抄録全体を表示

This paper describes a file-based workflow especially for receiving files in program delivery.

抄録全体を表示

In October 2011, NHK began operating high-definition (HD) Video Server System and News Center Archives System at its News Center. These two systems excel both in responsiveness and availability. Content in a file format and dramatic improvements in its affinity with metadata have transformed the workflow for news production.

抄録全体を表示

The AMWA-EBU FIMS (Framework for Interoperable Media Services) Task Force was established in December 2009. FIMS is a standardization activity of media industry common interfaces using a Service-Oriented Architecture (SOA) approach. This paper introduces the standardization activity, the overview of FIMS 1.0 specification published in April 2012.

抄録全体を表示

UMID is a SMPTE standard globally unique AV material identifier. This paper introduces the UMID and its applications with an emphasis of the need for the UMID application principles and the UMID resolution protocol as supplemental industry common rules. A new SMPTE project for them is also introduced with its future action plan.

抄録全体を表示

We developed the Flexible Program Production System, which is a new web-based system utilizing cloud computing technology, and released the source code. This paper describes the outline of the component software.

抄録全体を表示

This paper describes the situation of generating metadata for TV program and its materials in the broadcasting center. Automatization of the task and related trends are also described with the latest international standard for metadata.

抄録全体を表示

For reduction of power consumption in home,control of temperatures in room is one of the principal methods. We considered the experimental LED planter to be a small home and measured temperatures of air and circulating water with an AVR microcomputer. By using air fans and thermo heaters, the environment of the planter was controlled appropriate temperature for plants.

抄録全体を表示

We propose the information notification system using TV. When the detector detects the unusual event, it notifies TV of the location of the related information, and TV gets and shows it automatically.

抄録全体を表示

We have implemented a VNC (Virtual Network Computing) viewer into our digital TV sets. VNC has been widely used for PCs in order to enable us to view a screen of a remote PC connected through a network. Since most digital TVs in the market are equipped with a network interface and their embedded CPU becomes powerful enough to do some tasks in addition to its main jobs to control the TV operations, a VNC viewer can be implemented without adding or modifying the hardware. We have introduced a window system in order to allow users to change the size and the position of a VNC viewer screen freely within the whole TV screen.

抄録全体を表示

We developed System for writing Machines Unique Data(for example, digital key data) . This System ensures the confidentiality of data and improves production efficiency.

抄録全体を表示

Our recordable LCD TVs are equipped with front loading Blu-ray disc drive. The front loading provides ease of use for beginners, but it increases a size of packing box of the TV. This paper proposes a new mounting arrangement of the disc drive that can reduce the size of packing and presents its effect.

抄録全体を表示

We have focused on new magnetic memories without mechanical moving parts, utilizing the current-driven magnetic domain wall motions, in order to achieve ultra high data transfer rate. We fabricated [Co/Pd] magnetic nanowire with perpendicular anisotropy, and succeeded to drive multiple magnetic domains along the nanowire by the applied current.

抄録全体を表示

Hologram distortion can be compensated by controlling wavefront of the reference beam while reproducing. However, some wavefronts degrade high resolution bit data. We obtained wavefornt to compensate for hologram distortion without degradation of bit data by limiting moving distance of a deformable mirror. As a result, a bit error rate of the reproduce data could be decreased.

抄録全体を表示

A holographic memory is a promising candidate for next-generation optical memory. We implemented a twodimensional multistage equalizer that is composed of convolution filters including finite impulse response filter and interpolator. We present the structure of the equalizer and the experimental results that we evaluated bit-error-rate using the equalizer.

抄録全体を表示

We studied the readout characteristics of optical modulation on super-RENS disc that is one candidate of the next-generation high-density optical disc. The readout power to achieve a super-resolution effect can be decreased by increasing the modulation amplitude. The result also indicates the possibility to decrease the damage of the disc.

抄録全体を表示

This paper proposes a real-time protocol conversion method that can be provided the file-based storage for Ultra high Definition Television (UHDTV) systems. To verify the proposal, we developed a prototype system, making conversions UHDTV video signals to and from multiple MXF files.

抄録全体を表示

This paper describes the transform coding with orthogonal transforms which is derived from DCT. These transforms are switched due to the relative positions between a prediction unit and a transform unit. The performance is evaluated with HEVC Test Model (HM) ver. 6.0. The results showed 1.3 % bit saving on average than the original coding scheme.

抄録全体を表示

Transcoding from the existing video coding standard to HEVC (High Efficiency Video Coding) will become one of the most important technology for the future video services. However, complexity reduction approach for HEVC encoding is required to realize the video transcoder with the practical speed, because of the huge computational cost of HEVC. In this paper, we propose a low-complexity decision mechanism of prediction unit size especially for video transcoding from H.264/AVC to HEVC.

抄録全体を表示

We propose an objective video quality metric based on Hybrid NR framework to evaluate the pictures coded by HEVC. The proposed metric extracts three image features to estimate subjective quality: QP from the compressed bitstream and blockiness and flickering features from the decoded pictures. Computer simulations show that the proposed metric can estimate the MOS of HEVC-coded sequences at a correlation coefficient of 0.86.

抄録全体を表示

High Efficiency Video Coding (HEVC) realizes the adaptation to image characteristics by using the coding tree block structure. In this paper, we propose the intra prediction method in order to improve the coding efficiency. The proposed method allows splitting a coding unit by 1:3 ratios, and it also utilizes non-power-of-two size as the unit of transformation. The gain is 2.0% compared with the HM5.0 as an anchor method through an experiment.

抄録全体を表示

This paper provides coding performance comparison between High efficiency video coding (HEVC) and AVC/H.264 at Super Hi-Vision images (7680 x 4320 pixels). Coding results of HEVC exhibited coding gains of up to 63.8% from 22.2% compared to AVC/H.264.

抄録全体を表示

In this paper, we propose a method to decide the optimal viewpoint automatically when three-dimensional shape is projected to two-dimensional space. The optimal viewpoint in this paper is the viewpoint to be easy to recognize the shape. The method locates the optimal viewpoint using shape information and texture data.

抄録全体を表示

There are many rendering techniques using point-based super-sampling. By the super-sampling, more sampling gets more quality. But how many samples do we need? We can only estimate this from statistics. In this research, we introduce a new concept Micro point. Using the Micro point, we get aliasing free images without super-sampling.

抄録全体を表示

In this study, a new three-dimensional modeling technique using the implicit surface representation is proposed. In our approach, the implicit surface is expressed as a set of B-spline functions which enables high-speed rendering and intuitive modeling.

抄録全体を表示

We have estimated scattering properties of participating media using rendered images. We have rendered multiple scattering in participating media using photon mapping. Then, we estimate the scattering properties by comparing the generated images and real images.

抄録全体を表示

This paper proposes an enhanced motion graph construction method to automatically synthesize a wide variety of dance motions synchronized with music. Our motion graph has new metadata representing a contribution to the motion variety, which is estimated based on the variation of the steady-state transition distribution in partially modified the graph structure. Experimental results show that the proposed method increases the variety of synthesized dance motions.

抄録全体を表示